Fix error on ruby 2.5

Fix error on ruby 2.5

[Jeremy Evans]

Example code:

require 'socket'
require 'kgio'
addr = Socket.pack_sockaddr_in(80, 'www')
socket = Kgio::Socket.connect(addr)
buf = String.new(:capacity=>16384); nil
socket.kgio_read!(16384, buf)

Result on ruby 2.5:

TypeError: can't convert FalseClass into time interval

Cause:

Misuse of rb_funcall.  Not sure why ruby 2.5 is pickier than previous
versions.

Noticed when testing unicorn with ruby 2.5.

Fix:

--- ext/kgio/wait.c.orig
+++ ext/kgio/wait.c
@@ -92,12 +92,12 @@ static VALUE kgio_wait_writable(int argc, VALUE *argv,
 
 VALUE kgio_call_wait_writable(VALUE io)
 {
-	return rb_funcall(io, id_wait_wr, 0, 0);
+	return rb_funcall(io, id_wait_wr, 0);
 }
 
 VALUE kgio_call_wait_readable(VALUE io)
 {
-	return rb_funcall(io, id_wait_rd, 0, 0);
+	return rb_funcall(io, id_wait_rd, 0);
 }
 
 void init_kgio_wait(void)

[PATCH] wait: avoid passing unnecessary args to rb_funcall

[Eric Wong]

Jeremy Evans <code@jeremyevans.net> wrote:
> Example code:
> 
> require 'socket'
> require 'kgio'
> addr = Socket.pack_sockaddr_in(80, 'www')
> socket = Kgio::Socket.connect(addr)
> buf = String.new(:capacity=>16384); nil
> socket.kgio_read!(16384, buf)
> 
> Result on ruby 2.5:
> 
> TypeError: can't convert FalseClass into time interval

Odd, I haven't seen this; but using a fairly old gcc, though
(4.9.2-10 on Debian).

> Cause:
> 
> Misuse of rb_funcall.  Not sure why ruby 2.5 is pickier than previous
> versions.

Probably:

	r58362 ruby.h: check argc to rb_funcall
	r59483 ruby.h: NULL as empty array

?

Did you use the same compiler for previous Rubies?

> Noticed when testing unicorn with ruby 2.5.
> 
> Fix:

Looks good to me and I'll definitely accept it.

However, it is worrying to have rb_funcall suddenly break...

Anyways, I'll credit you as the author; but wanted to make sure
you're alright with the proposed commit message:

------------8<------------
From: Jeremy Evans <code@jeremyevans.net>
Subject: [PATCH] wait: avoid passing unnecessary args to rb_funcall

The extra arg is unnecessary and causes errors on some systems
under Ruby 2.5.0dev.  Note: this may be a regression in Ruby
2.5.0dev itself and might need to be fixed upstream, as other
gems may also break.

cf. https://bogomips.org/kgio-public/20171215053434.GT7579@jeremyevans.local/t/

Acked-by: Eric Wong <e@80x24.org>
[ew: commit message]
---
 ext/kgio/wait.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/ext/kgio/wait.c b/ext/kgio/wait.c
index 7bdf0a1..fa2fb4a 100644
--- a/ext/kgio/wait.c
+++ b/ext/kgio/wait.c
@@ -92,12 +92,12 @@ static VALUE kgio_wait_writable(int argc, VALUE *argv, VALUE self)
 
 VALUE kgio_call_wait_writable(VALUE io)
 {
-	return rb_funcall(io, id_wait_wr, 0, 0);
+	return rb_funcall(io, id_wait_wr, 0);
 }
 
 VALUE kgio_call_wait_readable(VALUE io)
 {
-	return rb_funcall(io, id_wait_rd, 0, 0);
+	return rb_funcall(io, id_wait_rd, 0);
 }
 
 void init_kgio_wait(void)
-- 
EW

Re: [PATCH] wait: avoid passing unnecessary args to rb_funcall

[Jeremy Evans]

On 12/15 06:32, Eric Wong wrote:
> Jeremy Evans <code@jeremyevans.net> wrote:
> > Example code:
> > 
> > require 'socket'
> > require 'kgio'
> > addr = Socket.pack_sockaddr_in(80, 'www')
> > socket = Kgio::Socket.connect(addr)
> > buf = String.new(:capacity=>16384); nil
> > socket.kgio_read!(16384, buf)
> > 
> > Result on ruby 2.5:
> > 
> > TypeError: can't convert FalseClass into time interval
> 
> Odd, I haven't seen this; but using a fairly old gcc, though
> (4.9.2-10 on Debian).

I'm using clang 5.0.0 (current OpenBSD/amd64 default compiler).

> > Cause:
> > 
> > Misuse of rb_funcall.  Not sure why ruby 2.5 is pickier than previous
> > versions.
> 
> Probably:
> 
> 	r58362 ruby.h: check argc to rb_funcall
> 	r59483 ruby.h: NULL as empty array
> 
> ?
> 
> Did you use the same compiler for previous Rubies?
 
Yes.  Older versions seem not have have problems, I regularly test
1.9.3-2.4.

> > Noticed when testing unicorn with ruby 2.5.
> > 
> > Fix:
> 
> Looks good to me and I'll definitely accept it.
> 
> However, it is worrying to have rb_funcall suddenly break...

Agreed, but in this case it seems like we just got lucky it worked in
earlier versions. I first noticed the error when testing with
2.5.0-preview1, but assumed it was a ruby regression that would be
fixed before the final release.  When it reoccured during testing of
2.5.0-rc1 is when I looked into the issue in more detail and found the
cause.

> Anyways, I'll credit you as the author; but wanted to make sure
> you're alright with the proposed commit message:

Yes, looks good to me.

Thanks,
Jeremy

> 
> ------------8<------------
> From: Jeremy Evans <code@jeremyevans.net>
> Subject: [PATCH] wait: avoid passing unnecessary args to rb_funcall
> 
> The extra arg is unnecessary and causes errors on some systems
> under Ruby 2.5.0dev.  Note: this may be a regression in Ruby
> 2.5.0dev itself and might need to be fixed upstream, as other
> gems may also break.
> 
> cf. https://bogomips.org/kgio-public/20171215053434.GT7579@jeremyevans.local/t/
> 
> Acked-by: Eric Wong <e@80x24.org>
> [ew: commit message]
> ---
>  ext/kgio/wait.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/ext/kgio/wait.c b/ext/kgio/wait.c
> index 7bdf0a1..fa2fb4a 100644
> --- a/ext/kgio/wait.c
> +++ b/ext/kgio/wait.c
> @@ -92,12 +92,12 @@ static VALUE kgio_wait_writable(int argc, VALUE *argv, VALUE self)
>  
>  VALUE kgio_call_wait_writable(VALUE io)
>  {
> -	return rb_funcall(io, id_wait_wr, 0, 0);
> +	return rb_funcall(io, id_wait_wr, 0);
>  }
>  
>  VALUE kgio_call_wait_readable(VALUE io)
>  {
> -	return rb_funcall(io, id_wait_rd, 0, 0);
> +	return rb_funcall(io, id_wait_rd, 0);
>  }
>  
>  void init_kgio_wait(void)
> -- 
> EW